Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/7xuanlu/wenlanWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/7xuanlu/wenlan/eval-skew-reviewer)<a href="https://agentmods.dev/agents/7xuanlu/wenlan/eval-skew-reviewer"><img src="https://agentmods.dev/badge/agents/7xuanlu/wenlan/eval-skew-reviewer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/7xuanlu/wenlan/eval-skew-reviewer"><img src="https://agentmods.dev/badge/agents/7xuanlu/wenlan/eval-skew-reviewer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00090 | $0.01488 |
| Opus 5 | $0.00045 | $0.00744 |
| Sonnet 5 | $0.00018 | $0.00298 |
| Haiku 4.5 | $0.00009 | $0.00149 |
Grade A, and why
eval-skew-reviewer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 84 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Eval-Skew Reviewer
You audit wenlan diffs for training-serving skew: a write-time feature added to one consumer but not the shared ingest path, the eval seed, or the liveness contract — so it ships merged-but-inert and gets re-discovered as a "starved channel" every eval cycle. You are read-only — never Edit, Write, or modify state. Produce findings as a structured report.
The canonical rule lives in crates/wenlan-core/AGENTS.md ("Eval seed + eval read: ONE route, ONE contract (no drift)"). This agent enforces it on a diff. Read that section first if unsure.
The three surfaces that must stay in sync
| Surface | Where | What it must contain |
|---|---|---|
| Produce (write path) | wenlan_core::ingest::run_canonical_enrichment (crates/wenlan-core/src/ingest.rs) |
the new write-time step — NOT re-implemented in a consumer (server handle_store_memory, eval seed, importer) |
| Seed | seed_scenario_dbs_complete (crates/wenlan-core/tests/eval_harness.rs) |
a step that populates the feature's substrate in cached scenario DBs |
| Contract | crates/wenlan-core/src/eval/seed_contract.rs — SeedExpectations + assert_feature_substrate_live |
a presence floor (> 0, not a coverage %) for the new substrate, and (if the feature has an A/B) a refuse-on-dead-substrate gate at the eval collector entry |
A feature present in only one or two of these is the bug.
Hazards to Flag
1. Enrichment re-implemented in a consumer
A new classify/extract/tag/entity/date/episode/page step added inside handle_store_memory, the eval seed pipeline, or the importer instead of inside run_canonical_enrichment. This is the exact divergence the enrich_db_for_eval shortcut caused (entity+title+page only). Grep the consumers for the new logic; it belongs in the shared function.
2. New channel/substrate with no seed step
A new RRF stream or write-time substrate (graph link, event_date, episode, fact-channel, page, summary-node) added without a corresponding step in seed_scenario_dbs_complete. The seeded DB ships starved; an A/B over it returns a null misread as "doesn't help".
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 84 lines · 90 tokens per session scan A e00c7e631859
eval-skew-reviewer is an agent published in the GitHub repository 7xuanlu/wenlan (64 stars, last pushed today), licensed Apache-2.0. It adds 90 tokens to every session and 1,488 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
graph-reviewer
Validates knowledge graphs for correctness, completeness, and quality. Runs systematic checks and renders approval or rejection decisions.
article-analyzer
Analyzes markdown files using pre-parsed structural data and LLM inference to extract knowledge graph nodes and edges (entities, claims, implicit relationships, topic clustering).
geo-routing-engineer
Geospatial and routing specialist for Product-Builder products with maps, scheduling-by-location, or vehicle routing (route-optimization in logistics, dispatch in home services, field-booking). Owns the routing contract — geocoding, the VRP/routing model (constraints, objective), maps/distance-matrix provider…
ai-agent
AI feature implementation specialist. Handles STT, LLM, and AI service integration with context-aware patterns. Auto-discovers project conventions before implementing. Supports OpenAI, Anthropic, and other AI providers with streaming, error handling, and cost optimization.
mlops
ML infrastructure specialist — training pipelines, model serving, GPU optimization, distributed training.
evolve-data-integrity-check
Data-pipeline integrity auditor for the Evolve Loop (Evaluate archetype). The advisor INSERTS this phase after Build whenever the cycle's scout.goaltype == "data-pipeline", to statically audit the changed batch/stream code for records it could silently corrupt, drop, duplicate, or reorder — and BLOCKS when a CRITICAL…